[QE-users] Enormous stress from a previously flawless input

[Andrii Shyichuk]

Dear Community,

Some time ago I've mentioned a potential problem: the input files that 
used to work fine (vc-relax) resulted in very high stress and no 
convergence in cell size.

I keep getting the same (or at least related) issue in PW QE 6.7, 
Intel+MKL.

In PW 6.7, the exemplary stress looks like this:

        316.18        0.00        0.00    ;
           0.00       32.20       -0.78   ;
           0.00       -0.78       33.     ;

I'm not sure if my formatting persists, hence the semicolons.

There is always a row 1 column 1 (cartesian x?) component of the stress 
which is about an order of magnitude larger than the second largest 
component.
Also, convergence in relax is never reached, total force remains rather 
constant.

This is an open-shell systems, a Lu2O3 cell with a single dopant atom at 
Lu site.

Every time this problem occurs - I have an electron in the system that 
does not want to localize at the dopant site and remains in conduction 
band. That might be physical or not - but it should result in some 
converged geometry, right?

Given the fact that this problem comes and goes - it might be related to 
hardware or an external library.

Where do I look for the source of it? Which external library could be 
corrupting data?
Or is it not an external problem?

Thank you.
Andrii Shyichuk, University of Wrocław.

> Dear Community,
> 
> I've encountered a weird behaviour of several vc-relax calculations.
> Namely, the SCF part is fine, but the stress part gives something like 
> this:
> 
>        total   stress  (Ry/bohr**3)                   (kbar)     P=   
> -35982.30
>   -0.24459504   0.00000000   0.00000000       -35981.17        0.00     
>    0.00
>    0.00000000  -0.24461371   0.00002601            0.00   -35983.91     
>    3.83
>    0.00000000   0.00002601  -0.24459958            0.00        3.83   
> -35981.83
> 
> As the steps go, total force remains about 0.25-0.3 and does not go 
> down.
> 
> The best part: I also got that on the inputs that used to work before,
> with the same version on PW.
> 
> My QE is 6.5, the release version, compiled with Intel 16 and MKL.
> 
> I am almost positive that this is a supercomputer (or even a hardware) 
> issue:
> - it came out of the the blue;
> - recompilation does not help.
> 
> The question:
> Did anyone ever encounter anything like that, and what that can be 
> related to?
> 
> Thank you in advance.
> Best regards.
> Andrii Shyichuk, University of Wrocław.